How to develop enough European AI startups

Jacques Bughin 26 February 2019

a

A

Europe may risk falling further behind the US and China, the leaders on the adoption and supply of artificial intelligence (AI). Both are investing aggressively in these technologies. Instead, Europe may find itself competing in AI with a rapidly expanding field that includes Canada, Japan, and South Korea, each of which is making strides in AI (CIFAR 2017, AIST 2017, Zastrow 2016). There is some evidence, though – notably the strength of Europe’s AI startups and depth of AI talent – that suggests that this view is too pessimistic. 

Europe’s AI gap 

The pessimistic view of European AI is based on several observations:

  • The digital goods and services balance with the US of each of the EU28 countries has been systematically negative, unlike traditional goods and services (Gomez-Herrera and Martens 2014).
  • As of the end of 2017, Europe was not home to any of the world’s 10 largest internet companies. In February 2017, Europe had only 10% of the world’s 185 unicorns – private companies with a value of at least $1 billion – compared with 54% for the US. China had 23% of unicorns (McKinsey Global Institute 2019). Only four European companies were in the top 100 global AI startups: Onfido and Tractable in the UK, Shift Technology in France, and Sherpa from Spain (CB Insights 2017).
  • Despite the fact that Europe has been a pioneer in testing and developing AI technologies, capital invested for digital startups has been subscale compared with the US and China. As of the end of 2017, the US has invested around €220 per capita in the US. In Europe, Sweden invested €123 per capita (the highest in the region) and Finland €58, but per capita investment was only €3 in Italy. In the provision of AI, Europe attracted only 11% of global venture capital and corporate funding in 2016. At this time, 50% went to US companies, with the balance going to Asia – mostly China (MGI 2017). 
  • In 2018, Europe had still not caught up (CB Insights 2017). In that year, China attracted almost half of global investment in AI startups, ahead of the US with 38%. The European Commission has announced a €2.6 billion investment in AI and robotics, but compare this with China spending €1.9 billion on a single AI technology park in Beijing (Larson 2018).

Determinants of AI startups – a production function approach

This narrative may be misleading. Investment in AI is only one input into the development of startups, of which Europe has many (Craglia 2018). Another input is human capital, and Europe has top university centres of AI and computer science, especially in France, Germany, Switzerland, and the UK. Consider, too, that Europe has 5.7 million professional software developers, compared with 4.4 million in the US (Atomico 2018). 

We lack, however, objective analyses of the factors – and their relative weights – that determine the dynamics of AI startups. Our own systematic analysis, alongside other research on US and European startup hubs (e.g. Valuer.ai 2019), led us to settle on a production function-like model that links AI startup density at the city and country levels with a set of production factors, and macroeconomic enablers that can shift the production frontier (MGI 2019). Our main findings include:

  • There are many other hubs than the over-used cases of Silicon Valley and Boston in the US. Europe has important hubs, too. Stockholm is one of the most prolific technology hubs in the world in per capita terms, home to 250 deep tech firms, up to 30% of which were AI-based startups in 2013. It has produced a remarkable number of new digital natives and unicorns. Helsinki is home to internationally known companies including Rovio, Supercell, and Linux, and today has a higher per capita startup density than Boston, New York, or Seattle in the US.
  • A common feature of successful AI hubs is that they are cities that have developed partnerships between public and private companies, and major universities. In the case of US hubs, two of the most notable universities are Stanford and Harvard. Entrepreneurs from these two hubs have founded many more unicorns (more than 50 from Stanford and close to 40 from Harvard) (Kotsch 2017). Paris has MINES ParisTech (Mines ParisTech 2018). Zurich has ETH, and Lausanne has EPFL (MGI 2018). 

Cities also tend to bet on a particular technology portfolio – for instance, Helsinki is not only strong in AI, but also in biotech and gaming (Vauer.ai 2018). Finally, public institutions play a role, and not only as a source of funding. Governments, for instance, can act as committed customer to secure enough demand to fuel AI startups. One example is e-government in Baltic countries. China’s government is also the main customer of AI ventures linked to video recognition.

  • We estimated a general translog function of AI startup density with 'traditional' production factors such as human capital (professional developers per capita) and AI capital (cumulative amount of deep-tech investment), augmented by other possible shift factors such as the amount of digital private-public partnerships, the scope of technology portfolios, and the ability to build innovative business models. Broadly, the model confirms that financing plays a material role: the elasticity is more than 0.7, double the 0.3 elasticity linked to the density of professional developers.

There are small complementarities between capital and labour, as anticipated and empirically found in academic work (Aghion et al. 2017 and Tambe 2014 are examples), which are large enough to create economies of scale at the level of cities and countries. We found that the ability to build innovative business models was statistically significant and a major correlate of AI startup density, with an estimated elasticity of 0.3 – as large as human capital. 

These analyses confirm that the amount of finance investment matters for the density of startups. Indeed, the US invested double the capital in digital startups as Sweden between 2015 and 2017 – and Sweden’s investment was double that of Finland’s. However, they also indicate that, to achieve scale in startups, investment combines with highly skilled developers and a culture of innovation. Cities with proven ability to innovate business models include Amsterdam, Tallinn, Stockholm, and Helsinki. 

Europe is not necessarily condemned to lag behind in AI, and may even thrive. Capital may flow sooner rather than later, and the European Commission may be able to play a role in overcoming today’s fragmented European development of AI, connecting different promising hubs in Europe to create scale.

References

Aghion, P, B F Jones, and C I Jones (2017), Artificial Intelligence and Economic Growth, NBER working paper 23928.

AIST (2017), Artificial Intelligence Research Center, Government of Japan.

Atomico (2018), "The state of European tech".

CB Insights (2017), "AI 100: The artificial intelligence startups redefining industries",

Craglia, M, (ed) (2018), Artificial Intelligence: A European Perspective, European Commission.

Gomez-Herrera, E and B Martens (2014), “The drivers and impediments for cross-border e-commerce in the EU”, Information Economics and Policy 28(c): 83-96.

Kotsch, C (2017), Which Factors Determine the Success or Failure or Startup Companies? A Startup Ecosystem Analysis of Hungary, Germany, and the US, Anchor Academic Publishing.. 

Larson, C (2018), “China’s massive investment in artificial intelligence has an insidious downside”, Science, 8 February.

McKinsey Global Institute (2017), Artificial intelligence: The next digital frontier?

McKinsey Global Institute (2019), Notes from the AI frontier: Tackling Europe’s gap in digital and AI.

McKinsey Global Institute (2018), The future of work: Switzerland’s digital opportunity

MINES ParisTech (2018), "Mines ParisTech launches a Post Master’s Degree in Artificial Intelligence and Movement with 4 European partners", Press release, 30 May. 

CIFAR (2017), "Pan-Canadian Artificial Intelligence Strategy", press release. 

Valuer.ai (2019), "The Top 50 Best Startup Cities in the World", 5 February.

Tambe, P (2014), "Big data investment, skills, and firm value", Management Science 60(6): 1351-1616.

Zastrow, M (2016), “South Korea’s Nobel dream”, Nature 534(7605): 20-23.

a

A

Topics:  Industrial organisation Productivity and Innovation

Tags:  AI, investment, hubs, startups, e-government

Senior partner, McKinsey; Director, McKinsey Global Institute

Events

CEPR Policy Research